An accountant should build a Power BI client dashboard from a reconciled monthly export, define a few measures the client can act on, and test access before sharing it. Power BI Pro costs $14 per user per month, paid yearly. Copilot in Power BI needs a paid Fabric F2-or-higher or Power BI Premium P1-or-higher capacity with Copilot enabled; Pro alone does not qualify.
For a firm owner, the first decision is whether clients need an interactive service at all. A monthly PDF or spreadsheet may be easier to control when the figures change only after close. Power BI earns its place when the same validated data supports repeatable views, drill-downs or multiple client conversations. Copilot is a separate capacity and governance decision, not a prerequisite for a sound dashboard.
Sell a reporting promise the source can support
Write a one-sentence promise for each client dashboard: “At the fifth working day after month-end, show closed-period revenue, gross margin, receivables ageing and cash movement, with links back to agreed source reports.” That is clearer than “real-time financial insight”. It tells the partner when to review, the client what to expect and the analyst which files to gather. If the underlying ledger contains unposted bank items, a live visual can make unfinished accounts look official.
Ask the client which decisions the page should help with. A trading business may want to see whether margin dropped because sales mix changed or input costs rose. A professional firm may care about unpaid invoices and billable work. Define the owner of each measure: who answers when the client disputes it? Do not load every available field simply because a connector exposes it. The KPI dashboard planning guide explains the broader choice of measures; an accountancy practice must add sign-off, period and source traceability.
Put a compact data contract beside the engagement. It should name the accounting source, the export owner, the period, when the export is frozen, the transaction fields, the currency convention, excluded accounts, the adjustment policy and the person who approves publication. For example: “Revenue means posted sales excluding credit notes dated after the cut-off; comparison uses the same definition for the previous period.” If that wording is disputed, resolve it before designing a chart. A beautiful trend can be wrong because its numerator has changed.
Keep management reporting distinct from statutory or tax work. A dashboard can show a management estimate, but a client should not assume every card is a filed figure. Add a visible “period ended” label and an “approved on” date. Ask the qualified accountant to review any financial interpretation before it goes to a client. If the visual feeds a decision about financing, distributions or compliance, check the underlying records and seek appropriate professional advice.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowBuild one client page from a controlled monthly pack
Begin with a repeatable export of transactions and balances rather than connecting several raw systems at once. Preserve the original file in a restricted location. In Power Query, the transformation part of Power BI, explicitly set date, number and text types; strip blank header rows; standardise account codes; and flag unexpected values. Create a calendar table and a mapping from account codes to reporting categories. Keep the mapping outside the visual so an accountant can approve changes. Document every exclusion, such as internal transfers that should not count as revenue.
Use a semantic model, Power BI's named set of tables and measures, with fields that a client can understand. “Net revenue” should have one definition everywhere. Gross margin should calculate from the same signed revenue and direct-cost measures as the chart beneath it. A card that pulls a manually typed total while a trend uses the ledger can disagree after an adjustment. Do not hide the difference with a rounded number. Reconcile each measure to the source report at month and year-to-date level.
Consider an illustrative small manufacturer. Its closed monthly export shows $96,000 invoiced sales, $4,000 credit notes, $54,000 direct costs and $18,000 overheads. Net revenue is $92,000; gross profit is $38,000; gross margin is about 41.3%. The dashboard also shows $27,000 outstanding receivables, of which $8,000 is beyond the agreed due date. Those figures are a worked example, not a client story. The accountant checks the $92,000 against the signed trial balance, then checks that the $54,000 direct-cost grouping includes the right accounts.
Now suppose one $6,000 supplier bill is posted late and belongs in direct costs. Gross profit falls to $32,000 and gross margin to about 34.8%. A client seeing the earlier 41.3% without a publication date could make a poor stock decision. The proper response is to mark the previous page superseded, record the adjustment, refresh and publish the corrected period after review. The change log should show who approved it. “Auto-refresh succeeded” is not evidence that a late bill has been handled correctly.
A separate illustrative service client has uneven revenue recognition. Its invoice export shows $40,000 this month, while the management accounts recognise $31,000. If the dashboard's first card is called “Revenue” without specifying which basis it uses, the partner and client may each believe a different figure. Name the card “Invoiced this month” or use the approved management-account measure. A definition note prevents a repeated reconciliation argument.
Choose a sharing pattern before inviting clients
Draw the access path for each client: report author, internal reviewer, client viewer and administrator. Decide whether to create a separate workspace or report per client, or a shared semantic model with row-level security. Row-level security restricts rows by identity and role; Microsoft's documentation says it applies to Viewer users, not workspace Admin, Member or Contributor roles. That detail matters when a well-meaning staff member adds a client as a workspace Member to “fix access”. Test with a client-equivalent account, not only with an internal administrator account.
For five clients, separate workspaces may be simpler to inspect, although maintenance of repeated measures becomes harder. For fifty, a shared model with carefully maintained client IDs may be worth the setup. Neither route is safe without a permission test. A client code used in an export must match the identity mapping; a blank code should go to an exception list, not to an “all clients” bucket. Never use a public link for confidential accounts. Consider what a user can download or export as well as what the screen displays.
Try this illustrative access test: client A signs in and sees $92,000 revenue for their business. Search the report for client B's code, use available filters and export a table. The expected result is that client B's rows remain inaccessible. Repeat as an internal reviewer and as a client with a newly changed email address. A pass from an administrator's own view tells you very little. Record the test account, date, report version and observed results as part of the handover.
Power BI's free account can create reports, while its pricing page says Pro is used to publish and share. Pro is listed at $14 per user monthly, paid yearly; Premium Per User is listed at $24 per user monthly, paid yearly. Client viewing rights depend on the chosen sharing and capacity arrangement, so do not multiply $14 by an assumed audience until your administrator checks the exact route. For a three-person internal team, an illustrative starting list-price calculation is three Pro seats × $14 = $42 monthly, paid yearly, before any capacity or external sharing costs. It is a budget scenario, not a quote.
Check what Copilot actually needs
Microsoft's Copilot requirements state that a paid Fabric capacity of F2 or higher, or Power BI Premium capacity of P1 or higher, is required. A free or trial capacity does not qualify. The relevant admin setting must be enabled, and the user must have the correct access for the particular experience. The standalone, full-screen Copilot bills to a Fabric Copilot capacity if the organisation has one, or otherwise to an eligible workspace on F2 or above, so ask the administrator to check the exact configuration rather than assume every user needs a separate capacity. Microsoft states plainly that a Pro or Premium Per User licence alone is not sufficient, because Copilot needs organisational capacity. In Power BI Desktop, the author also needs write access to a workspace on paid Fabric or Premium capacity. After buying or scaling capacity, Microsoft warns that Copilot can take up to 24 hours to become available, so do not schedule a client demonstration for the same afternoon.
Copilot can help a report user ask questions about a model and can help an author draft a calculation or summary. It does not repair an ambiguous chart definition. If “margin” is a raw column in one table and a calculated ratio in another, a plausible answer may use the wrong one. Prepare names and descriptions in the model, hide technical columns from client-facing exploration, and test the likely questions against signed figures. Use the Power BI Copilot pricing guide when the firm is genuinely considering capacity; do not buy it merely because the button appears in a demonstration.
An illustrative prompt to test a finished model is: “Compare the approved gross margin for the last two closed months. State the source period and exclude draft postings.” A sample output might say, “Gross margin rose from 38% to 41%, mainly due to higher sales.” The accountant should then inspect the period filters and verify the two ratios. If a $6,000 late bill changes the second month, the sentence may be false or premature. Rewrite the output to state the verified change and the limitation: “The corrected margin was 34.8%; the earlier draft excluded a direct-cost bill.” Keep the original query and review decision if the summary is sent externally.
Another example is a client asking, “Which unpaid customers should I chase first?” A dashboard may rank invoice amounts and due dates, but a model cannot know that an invoice is under dispute unless the source records that status. The safe answer includes the data boundary: “These are the largest overdue balances in the current ledger; check disputes and payment promises before contacting anyone.” Add a dispute flag to the source if the question recurs. An AI narrative should not invent a collection recommendation from an incomplete ageing report.
Measure the refresh and reconciliation burden
Scheduled refresh is useful only if its credentials, gateway where needed, and source schedule are maintained. Microsoft's refresh guidance explains the service settings. Two details matter for a monthly client pack: scheduled refresh pauses after two months in which nobody views a report or dashboard built on the model, and the schedule switches itself off after four consecutive failures, for example when a stored password expires. Pro allows up to eight scheduled refreshes a day; Premium Per User and capacity allow up to 48. A dashboard viewed only at quarter-end can therefore stop refreshing between visits. Put the last successful data refresh time on the page, but keep it separate from the accounting period close date. A refresh at 09:00 may simply reload last month's unchanged export. Monitor failed refreshes and teach the team how to publish manually when the schedule fails.
Use a monthly control sheet with five lines: source balance, dashboard balance, variance, explanation and reviewer initials. For the manufacturer's $92,000 net revenue, the variance should be $0 after approved exclusions. If the dashboard shows $92,400, trace the extra $400 to a duplicated transaction, a refund treatment or a period filter before release. A rule such as “variance must be zero unless a named reconciling item is recorded” is more useful than an arbitrary green status. The dashboard accuracy checks give a wider test set, while this control ties the page to accounting evidence.
A fifth illustrative case concerns receivables ageing. One invoice of $2,500 is dated the last day of the prior month but was posted on the third day of the current month. If the visual ages by posting date, the balance may appear younger than the client expects. Document whether ageing uses invoice date, due date or posting date; compare the dashboard to the firm's approved ageing report. An attractive bar chart cannot resolve a date-definition dispute. The reviewer should sign off the definition before the client sees it.
Track the work required to keep the page reliable: export and cleanup minutes, reconciliation minutes, access administration, refresh failures and client questions. In an illustrative month, two hours of data preparation, one hour of reconciliation and one hour of partner review are four hours of service work, even if the software licence is only $42 in the internal-seat scenario. If the client needs only one static chart monthly, a spreadsheet may be a better service design. If several recurring decisions use the same approved data, that preparation can support a more useful dashboard.
Write change requests into the monthly release routine
Clients will ask for a new filter, a revised comparison or a figure from another system. Treat each as a change to the reporting contract. Record the requested decision, the extra data source, who can validate it, the access effect and whether earlier months need restating. A request to split sales by product may be straightforward if every ledger transaction carries a reliable product code. If half the historical entries lack one, a new chart would imply a precision the records do not have. Show “unclassified” explicitly and agree whether the missing codes are worth repairing.
For example, an illustrative manufacturer asks to see the margin of two product lines. The ledger has $92,000 net revenue, but $7,000 of sales and $3,000 of direct costs have no product code. A dashboard that simply drops the unclassified rows will make the two product-line margins look cleaner than the whole business. The accountant can publish total gross margin and a product split labelled “classified transactions only”, with the excluded $7,000 and $3,000 visible. If the client intends to close a product line based on that split, repair the coding first and review with the responsible accountant.
A second change request asks for cash forecast rather than historic cash movement. That needs assumptions about receipts, payment dates and commitments; it is not a different colour of the existing cash chart. Keep forecast values in a separate table with an as-of date, assumption owner and scenario label. Show actual cash beside forecast, and calculate variance when the next period closes. A model can chart the assumptions, but it should not present uncertain future receipts as bank balances. The client may prefer a short discussion of the assumptions to another dashboard page.
Use version notes when definitions change. “Receivables overdue” might originally include every invoice past its due date. If the client later asks to exclude disputed invoices, keep a dated note of the change and state whether earlier periods were recomputed. Without that note, a falling overdue balance could appear to be a collection success when it is just a filter change. Give the reviewer a preview showing old and new totals before publication, including at least one known disputed invoice to prove the filter works.
Finally, decide who answers a client query after publication. A responsible accountant should be able to trace a chart point to the source extract, mapping and reconciliation sheet without asking the dashboard maker to reverse-engineer it. Keep the approved source file, transformation version and release note together. If the client challenges the $8,000 overdue amount, the team should identify the invoices and cut-off used, then correct either the record or the explanation. That is the service standard the software has to support.
Agree on a release test before offering the dashboard as a standing client service. The test should cover two closed periods, a late adjustment, a client with no data, an identity change, a failed refresh and a dispute over one measure. The partner signs off the definitions and numbers; the client-facing owner checks wording and access; the administrator checks capacity and licences. Once these pass, use a small, clearly labelled page and add measures only when a client decision calls for them. The firm's professional judgement stays visible even when Copilot helps draft an explanation.
Dashboard purchase questions for a small firm
Does Power BI Pro include Copilot in Power BI?
A Pro licence lets an eligible user publish and share reports, but it does not by itself satisfy Copilot's capacity requirement. Microsoft states that Copilot needs access to paid Fabric F2 or higher or Power BI Premium P1 or higher, plus the relevant administrator setting. Check the particular Copilot experience and capacity assignment before budgeting.
Can a client see only their own figures in a shared model?
Power BI row-level security can restrict rows for users with Viewer permissions, but it does not apply to workspace Admin, Member or Contributor roles. A firm should test each client identity as a viewer and verify exports and links. For a small number of clients, separate workspaces or reports may be easier to audit than a complicated shared model.
How frequently should an accounting dashboard refresh?
Match refresh to the source and the promise made to the client. If the ledger is closed monthly, refreshing the visual every hour will not make the figures current. Label the reporting period, source extraction time and last successful refresh separately. Test one closed period and one late adjustment before enabling scheduled refresh.
Further reads
- Power Automate for Accountants: Flows for Documents and Approvals — Route source documents and preserve sign-off.
- ChatGPT for Accounting With Excel: Seven Tasks, Each With Its Check — Keep spreadsheet-based accounting checks separate.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Microsoft Power BI pricing; Microsoft Learn, Copilot for Power BI overview and requirements; Microsoft Learn, Row-level security with Power BI; Microsoft Learn, Configure scheduled refresh (checked 4 November 2026).